can a circular payment card format effectively elicit ... · in particular, the format can...
TRANSCRIPT
Can a Circular Payment Card Format Effectively Elicit Preferences? Evidence from a Survey on a Mandatory Health Insurance Scheme in Tunisia
Olivier Chanel Khaled Makhloufi
Mohammad Abu-Zaineh
Working Paper No. 02/2016
AHEAD WORKING PAPER SERIES/DOCUMENT DE TRAVAIL
1
Can a circular payment card format effectively elicit preferences?
Evidence from a survey on a mandatory health insurance scheme in Tunisia
Olivier Chanel
Aix-Marseille University (Aix-Marseille School of Economics), CNRS & EHESS, GREQAM
and IDEP, Centre de la Vieille Charité, 13236 Marseille Cedex 02, France
e-mail: [email protected]
Khaled Makhloufi
INSERM-IRD-UMR 912 (SESSTIM), Aix-Marseille University, 13006 Marseille, France
e-mail: [email protected]
Mohammad Abu-Zaineh
INSERM-IRD-UMR 912 (SESSTIM), Faculty of Medicine and Aix-Marseille School of
Economics (AMSE), Aix-Marseille University, 13006 Marseille, France
e-mail: [email protected]
∗Corresponding author: [email protected]
Keywords: Willingness to pay, elicitation format, payment card, circular payment card, social
health insurance.
JEL Codes: C93; I19;
2
Abstract:
The choice of an elicitation format constitutes a crucial but tricky aspect of stated preferences
surveys. It affects the quantity and quality of the information collected on respondents’
willingness-to-pay (WTP) as well as the potential errors/biases that prevent their true WTP
from being observed. Here, we propose a new elicitation mechanism, the circular payment card
(CPC), showing that it helps overcome the drawbacks of the standard payment card (PC) format.
Indeed, it uses a visual pie-chart representation without start or end points: respondents spin the
circular card in any direction until they find the section that best matches their true WTP.
We perform a contingent valuation survey regarding a mandatory health insurance scheme in a
middle-income country, Tunisia. Respondents are randomly split into three subgroups and their
WTP elicited using one of three formats: open-ended (OE), standard PC and the new CPC. We
compare the WTP observed and find significant differences in unconditional and conditional
analyses. Our empirical results consistently indicate that the OE and standard PC formats lead
to significantly lower WTP than the CPC format. Overall, our results are encouraging,
suggesting that further empirical studies could establish CPC as a valid alternative to existing
formats, especially for low- and middle-income countries.
Research highlights:
* We propose a new preference elicitation technique: the Circular Payment Card
* We compare its theoretical and procedural properties with existing techniques
* We test its feasibility on a survey dealing with a mandatory health insurance scheme
* We find it a useful alternative, especially for poor and less educated populations.
3
1. INTRODUCTION
Contingent valuation (CV) surveys reveal respondents’ true unobserved willingness-to-pay
(WTP) values, provided an appropriate elicitation format is chosen. Since respondents only rely
on this component and the hypothetical scenario when making their decision, elicitation can
affect the quality and quantity of WTP information collected, as well as potential errors/biases.
In particular, the format can introduce implied value cues that directly lead to respondents
anchoring on the proposed bids, taken as an indicator of the quality of the good (Boyle et al.
1997) or as plausible values (Herriges and Shogren 1996), even if the numbers proposed are
unrelated to the good (Wilson et al. 1996). Implied cues also include a tendency to answer “Yes”
(“yea-saying”) to avoid answering “No” (Kanninen 1995). None of the various existing
elicitation formats stands out from the rest: they provide different types of information, collect
different data on WTP (continuous, binary or interval) with its statistical and practical
properties (in particular, incentive compatibility), widely discussed in the literature (see Carson
and Groves 2011). As Shono et al. (2015) recently stated, “which of these methods is more
reliable in the context of low- and middle-income countries remains controversial” (p. 33).
This paper seeks to contribute to the literature on WTP elicitation formats by assessing whether
preferences can be elicited effectively by a new technique: the Circular Payment Card (CPC).
Unlike the standard Payment Card (PC), the new CPC format uses a visual pie-chart
representation without start or end points. Respondents spin the circular card in any direction
until they find the section that best matches their true WTP. Here, we test whether this CPC can
provide insights into the current debate on Universal Health Coverage (UHC) in Tunisia (Vega
2013; O’Connell et al. 2014). Overcoming the obstacles to UHC is important for governments
of low- and middle-income countries, if they hope to tackle health equity issues and allow for
health risk-sharing (Who, 2013). Our CV survey asks uninsured Tunisians their WTP for a
4
mandatory health insurance scheme. We use between-respondent analysis to compare the
WTPs elicited using the CPC with those elicited using Open-Ended (OE) and PC formats. Our
empirical results consistently indicate that the OE and standard PC formats lead to significantly
lower WTP than the CPC format.
The remainder of the paper is structured as follows. Section 2 presents the OE and PC elicitation
formats, compares their theoretical and procedural properties, and details the new format
proposed. Section 3 describes how the survey was administered, and the statistical and
econometric methodology. Section 4 reports the results, which section 5 discusses.
2. COMPARING ELICITATION FORMATS IN CV SURVEYS
Mitchell and Carson (1989) listed nine elicitation formats 25 years ago, but we only discuss the
five most used since then, along with their variants.1 Although the NOAA (Arrow et al. 1993)
recommended the Dichotomous Choice (DC) format, it has been criticized in the literature for
overestimating the mean true WTP2 as well as for practical reasons. First, it needs an ex ante
choice of bids to be proposed (and an inefficient set of bids may affect mean WTP; Kanninen
1995). Second, this elicitation technique carries implied value cues: anchoring, starting point
bias or “yea-saying”. Thus, it requires larger samples to provide enough information to
determine true WTP. Extending the single-bounded version to the double-bounded version, by
introducing a second bid conditional on the answer given to the first bid, improves efficiency
1 The other elicitation methods are the sealed-bid auction, the oral auction, the checklist (similar to payment card)
and the spending question offer. 2 Among the 21 CV studies comparing the DC format with OE and PC formats between 1990 and 2005 and
surveyed in Champ and Bishop (2006), none showed lower mean WTP with the former than with the latter two
(see also Lunander 1998; Welsh and Poe 1998; Cameron et al. 2002; Carson and Grooves 2007; Frör 2008;
Uehleke 2014). Hammerschmidt et al. (2003) is one exception, although on a small sample size (n=73) and for
one only of the three diabetic complications assessed in the survey.
5
but does not remove the implied value cues; it may therefore take large samples to reduce the
risk of poor information on WTP distribution.3 Bidding games, in which respondents face
several rounds of discrete choice questions involving increasing/decreasing bids (or both, see
for instance the “structured haggling technique” proposed by Onwujekwe, 2004), can induce
higher extreme responses in addition to implied value cues. These differences explain the lack
of consistency in results from different elicitation methods applied to the same dataset (see
Cameron et al. 2002). In the following, we choose to focus on the two last formats: OE and PC.
2.1. Comparing OE and PC
In the OE direct question, the respondent provides a monetary value that directly corresponds
to a change in utility. This can provide more accurate WTP values than other elicitation methods,
as it is not prone to implied value cues. Yet it may be difficult to answer, especially when the
purchasing decision involves unfamiliar (non-market) commodities. Moreover, it may lead to
a higher non-response rate and to more “Don’t know”s and extreme responses than other
formats.
In the standard PC (or payment ladder, proposed by Mitchell and Carson, 1981), respondents
are required to choose a bid as close as possible to their true WTP from a list with several bids.
Respondents answer or tick “Yes”, “No” or “Don’t know” for each bid. The PC has become
popular since it mimics real life decisions (deciding whether or not to buy a good at a given
price) better than the OE (deciding the maximum price one would pay for a good). In addition,
PC offers visual support that facilitates the construction of evaluation and avoids a high rate of
non-response and overestimated values. Although the method reduces the biases associated
with closed-ended and bidding formats (especially when starting with very low (or zero) values
3 Neither the one-and-a-half bound format (see Cooper et al. 2002) nor the triple-bounded format (see Bateman et
al. 2001), have been proven to provide substantial improvements (see Bateman et al. 2009).
6
and ending with very high values), biases due to starting values, the range and the centering of
bids may occur. Variants tackle starting-point bias by introducing randomness, like the
randomized card sorting first used by Carthy et al. (1999). Respondents sort a set of randomly
shuttled cards (with one bid on each card) into three piles: one they would pay for sure, one
they would definitely not pay and one they are uncertain about paying.
PC is really the only format that takes respondents’ uncertainty directly into account through
the WTP elicitation mechanism, and has been improved since 1981. Indeed, Dubourg et al.
(1994) propose a two-way payment ladder providing two WTPs: the maximum s/he is definitely
willing-to-pay and the minimum s/he is definitely not willing-to-pay. Welsh and Poe (1998)
proposed a multiple-bounded DC with five possible answers (“Definitely no”, “Probably no”,
“Not sure”, “Probably yes” and “Definitely yes”). Cook et al. (2012) used a traffic light analogy,
where green prices mean “sure to buy”, yellow prices “uncertain to buy” and red prices “sure
not to buy”. Further developments by Wang and Whittington (2005) used a stochastic card
approach that adds to the five possible answers numeric likelihood values (0%, 25%, 50%, 75%
and 100%). Recently, Mahieu et al. (2014) allowed respondents to choose the shape of their
uncertainty function (uniform, asymmetric or symmetric triangular).
Regarding empirical results, Whynes et al. (2005) compared five elicitation formats to
determine the WTP mass population screening for colorectal cancer: DC, bidding game, OE
and two PC formats. They obtained far higher values from the two former than from the three
latter formats for median WTP, elasticity and consumer surplus. Veisten and Navrud (2006)
applied an induced truth-telling mechanism to study variations in WTP to protect virgin forest
areas around Oslo against cutting, depending on the elicitation format used. In addition to an
overestimation of WTP with respect to actual payment, they found evidence that a greater
uncertainty effect and “yea saying” effect in DC led to higher WTP than with OE. Comparisons
between OE and PC have not proved conclusive: OE led to lower WTP than PC in Donaldson
7
et al. (1997) or Whynes et al. (2003) for health-related goods, but equality or weak evidence of
the reverse relationship was found in Lopez-Martin et al. (2008) for environmental goods and
Gyrd-Hansen et al. (2014) for an ambulance helicopter service.
2.2. Incentive compatibility of OE and PC
First and foremost, the incentive compatibility issue for direct revelation mechanisms only
makes sense for consequential surveys; i.e. “If a survey’s results are seen by the agent as
potentially influencing agency’s actions and the agent cares about the outcome of those actions”
(Carson et al. 2007; p.183). In a strict sense, incentive compatibility means that the properties
of the value elicitation format ensure that there is no advantage in answering strategically: the
truthful preference revelation is an optimal (and the dominant) strategy for the respondent –
referred to as Theoretical Incentive Compatible (TIC). In a CV sense, answers may also depend
on the psychological or psychometric properties of the elicitation format – referred to as
“Procedural Invariance” (PI).
The OE format has generally been considered as not TIC (see, e.g. Hoehn and Randall 1987):
according to Gibbard (1973) and Satterthwaite (1975), no format with more than two
alternatives can be TIC without additional restrictions on respondents’ preferences (Carson and
Grooves, 2007). However, Green et al. (1998) proved that it is possible to make an OE question
TIC if the agent thinks s/he has a positive probability of being decisive and if the payment
vehicle used is decoupled (see also Schläpfer and Bräuer 2007; for field experiments on this
topic) and Riera (2003) showed that the OE format can be TIC when the final decision relies
on a quantile of the respondent's WTP distribution (e.g., the median) rather than the mean. The
OE format is PI inasmuch as no potential information is given to respondents through bids, but
may lead to more zero WTPs when respondents think their true WTP is lower than the cost
8
anticipated if the good is provided (Carson and Grooves 2007).
The PC cannot be proven to be TIC and may, in rare cases, give rise to very high values.
However, Carson and Grooves (2011) conjecture that using a set of values (instead of one) may
do two things: turn it into a choice with a sufficiently large number of options to approximate
a continuous matching response; and increase uncertainty over the actual cost of the program,
which generally favors conservative values.
The PC format is not PI, being sensitive both to the order in which bids are proposed to
respondents, so that significantly higher WTP is elicited for the version starting with high values
than for the version starting with low values (see e.g. Dubourg et al. 1994, 1997; Smith 2006),
and to the range of bids (see Whynes et al. 2004; Covey et al. 2007). However, “yea-saying” is
less prevalent in the PC format, where respondents still express support for the proposed good
even if they choose a low WTP, as opposed to the DC format, where respondents may disregard
the level of the bid proposed in order to express their support (Uehleke 2014).
2.3. An improved PC format: the CPC
Despite recent improvements, there remain disadvantages to the PC format, chiefly the risk of
implied value cues from the range of the bid interval, the starting values and the position of the
bids. Improvements on uncertainty did not address this, and random card sorting shows very
mixed results. For instance, while Covey et al. (2007) found that the range bias can be at least
as great as with standard PC, Smith (2006) showed that random card sorting can produce the
most ‘valid’ PC format values, though he did not test for range bias.
To overcome these disadvantages, we propose a new form of PC: the Circular Payment Card
(CPC). Unlike the standard PC, the CPC relies on a visual representation of a circular card with
no predetermined start or end points, no top or bottom, no left or right (see Figure 1 in Appendix
9
2). The interviewer asks the respondent to think about his/her WTP, and then presents the sheet
with the printed CPC in a random position. Respondents are asked to spin it in the direction
they want, until they find the section that best corresponds to their WTP values. The text
containing the WTP values is curved around the circle to allow easy handling and spinning,
with no predetermined direction of rotation. 4
In addition to the advantages of the standard PC format, CPC eliminates starting-bid bias
(because each section is equally likely to be seen at first glance) and middle-card bias (by
construction). Following Carson and Groves (2011), it helps strongly reduce the range effect
associated with the bids chosen, as the succession of bid ranges mimics a continuous
distribution. The circular representation and the spinning reinforce this, and make CPC
indisputably more PI than PC. Moreover, spinning the circular card to reach the section
corresponding to the ex ante WTP chosen requires both cognitive and physical effort, contrary
to both the standard PC (increasingly high, easy-to-read bids proposed) and OE (no bid
proposed). This may encourage greater involvement during the elicitation process, especially
among socially disadvantaged and poorly educated populations. Finally, CPC helps reduce
survey costs by being compatible with mail surveys (inserted in the envelope with the survey
material) and Internet surveys (displayed on the screen in a random position; respondents spin
it as desired by pressing keyboard keys).
3. RESEARCH DESIGN
3.1 Study design
4 Note that Dubourg et al. (1994) tested for (and found) evidence of starting point bias by using a disk with a small
window revealing a single bid at a time, presenting half of the sample with a low starting bid and the other half
with a high starting bid. However, this differs crucially from our proposed CPC in which respondents are required
to spin the whole CPC until they find their maximum WTP values.
10
The questionnaire was developed by five researchers specialized in CV and stated preference
surveys. After two pre-tests of respectively 20 and 30 respondents, it was fine-tuned for range
and centering of bids (PC and CPC) and for wording (in Arabic). The questionnaire consists of
four parts. The first introduces the objectives of the CV study and collects general information
on the respondents’ demographic characteristics, their recent health experiences (health status
and utilization of healthcare) as well as the reasons for being excluded from the current health
insurance mechanisms. The second part describes the current formal “public single-provider
scheme” - run by the “Caisse Nationale d’Assurance Maladie” (CNAM). This scheme covers
healthcare services that are only provided by public sector facilities, with a cap on annual co-
payments (Makhloufi et al. 2015). Respondents are then asked to reveal their willingness-to-
join and their quarterly WTP for this Voluntary pre-payment Health Insurance Scheme
(VHIS).5 The last part collects information on the respondents’ socio-economic characteristics,
and their comments on the proposed schemes. To ensure comprehension of the questionnaire,
the benefits of the health insurance scheme were fully explained by the interviewers (see
Appendix 1). All subjects were asked for their full consent to participate in the study and no
financial incentives were offered.
3.2 Data collection
The CV study was conducted in Tunisia between August 1st and September 30th 2013. The
inclusion criterion was being a Tunisian citizen not covered by – nor benefiting from - any
health scheme (including the state-subsidized medical assistance programs). As shown
elsewhere (Abu-Zaineh et al. 2013; Makhloufi et al. 2015), the excluded segment of the
5 In the third part, respondents are asked to choose between two other mandatory insurance plans run by the CNAM,
and then to declare their maximum WTP for them. These WTP questions, asked subsequently, cannot influence
the initial VHIS elicitation process and we do not exploit them here.
11
population in Tunisia mainly consists of informal sector employees (Gatti et al. 2011) and the
unemployed (18.3%, NIS, 2011). Therefore, to ensure representativeness, two types of
sampling location were identified in the three main regions of Tunisia (North, Central and
South): (i) the “Souk”, where many informal activities take place and (ii) the “Al-mydan” (i.e.
public square), the site of many youth demonstrations since 2011 and the so-called “Jasmine
Revolution”.
We used face-to-face interviews by fully-trained interviewers for two reasons: the surveyed
population consists partly of informal employees and rural residents of a developing country
(Tunisia), and the survey is complex, comparing different insurance schemes (with questions
depending on previous choices and rankings). Of the initial sample of 456 subjects, 30 refused
to participate, giving a response rate of 93.42%. The overall sample was then randomly split
into three equal sub-groups. Each sub-group was randomly assigned to one elicitation format
(OE, PC or CPC) to answer the WTP question for VHIS (see Appendix 2). The same
questionnaire was used for all sub-groups, apart from the WTP elicitation format.
3.3 Methods
3.3.1 Descriptive unconditional analysis
The mean WTP obtained from the three elicitation formats (OE, PC and CPC) was compared
using a two-sample mean bi-lateral equality test (Ryan et al. 2004). In addition, we tested for
equality of median WTP across elicitation formats (Fisher's exact test for p-values) and used a
chi-squared test of differences in proportions to test for differences in distribution of the first
and the last interval of the payment cards (Champ and Bishop 2006). Lastly, since equality tests
are sensitive to distributional assumptions, we performed non-parametric and distribution-free
Kolmogorov-Smirnov tests (Chakravarti et al. 1967).
12
3.3.2 Conditional analysis
We examined differences in elicitation formats by controlling for the determinants of WTP.
Some respondents were directly asked for a point estimate of their WTP (OE format), while
others were offered an interval with two specified thresholds (PC and CPC formats). The
interval regression model (cf. Wooldridge 2002; Anderson and Mellor 2009) was chosen for its
ability to handle WTP elicited with all three formats in one single model: exact values in the
OE, the interval in the PC formats. All statistical analyses are performed using STATA® 12.1
software and use heteroscedastic-robust standard errors.
4. RESULTS
4.1. Descriptive Results
Table 1 provides descriptive statistics on WTP in addition to respondents’ socio-economic,
socio-demographic and health characteristics (health status, healthcare utilization and capacity
to pay for a given illness). We find significant differences across elicitation formats for several
variables (details upon request): High School (p-value=0.0157), Income (p-value=0.0202),
Equivalised Income (p-value=0.0425), Rural (p-value=0.06), Disadvantaged Governorate (p-
value<0.001) and Outpatient Respondent (p-value=0.0806). Although the elicitation formats
are perfectly distributed across interviewers (p-value=0.980), the latter are strongly correlated
with Rural, Disadvantaged Governorate and Public Squares. This is because only 5 interviewers
covered the three regions of Tunisia studied, generating strongly unbalanced distributions of
the characteristics of the 8 sample locations by interviewer. Interviewers’ dummies and these
three spatially related variables will therefore not enter econometric models simultaneously,
due to high collinearity.
13
On average, respondents belong to households earning about one and a half times the minimum
monthly salary in Tunisia (TND 558.116). Interestingly, 49% of respondents reported a lack of
insurance coverage due to their professional activities not being officially declared, in line with
previous studies suggesting that almost half of Tunisian youth work in the informal sector
(Loayza and Wada 2009; Gatti et al. 2011).
[PLEASE INSERT TABLE 1 ABOUT HERE]
4.2. Comparing formats: Unconditional analyses
Table 2 presents descriptive statistics for the three elicitation methods. Our first question is
whether more comprehension or completion difficulties arise with the CPC format than with
the two others. Equality tests show that CPC does not entail either longer response times or
greater difficulties answering WTP questions than OE or PC (see upper part of Table 2).
Second, regarding WTP,7 the mean for the CPC (WTPCPC) appears to be consistently higher
than with the other two formats. Interestingly, using the CPC also yields the highest median
WTP (TND 45), while the PC format yields the lowest median (TND 35), with the OE format
in between (TND 40).
[PLEASE INSERT TABLE 2 ABOUT HERE]
We test for equality among mean and median WTPs and compute the corresponding p-values.
Results reported in the middle part of Table 2 show that both mean and median WTPCPC are
6 At the time of the survey, 1 Tunisian Dinar (TND) = € 0.455 = $ 0.605. 7 In this section, we approximate WTP for the two PC formats with the middle of the bid-range elicited (Cameron
and Huppert 1989; Yang et al. 2012). The conditional analyses will use the actual lower and upper bounds for
these formats.
14
significantly higher than those of WTPPC and WTPOE. No significant differences are found
between the mean and median of the two latter.
[PLEASE INSERT TABLE 2 ABOUT HERE]
Figure 2 shows the whole distribution of WTP by elicitation method. Mean and median WTP
roughly summarize overall WTP distribution, and do not inform on extreme values. The OE
format is assumed to provide more extreme responses (i.e. lower and higher WTP) than other
formats, and the PC to be subject to starting point anchoring. This is tested through pairwise
equality of proportion tests on the proportion of answers corresponding to the first (0-10) and
the last (more than 130) bid-ranges. Results in the lower part of Table 2 show that the OE
elicitation format leads to a significantly higher proportion of low WTP than the CPC (p-
value=0.012), but there are no significant differences for high WTP (p-value=0.3496).
[PLEASE INSERT FIGURE 2 ABOUT HERE]
Figure 3 presents the proportion of accepted bids by elicitation method. As shown, the CPC
seems to lead to a higher probability of acceptance, which means that the whole distribution of
WTP is influenced in an upward direction, not only mean and median WTP. Distribution-free
Kolmogorov-Smirnov tests confirm that the OE and the PC elicitation formats do not lead to
significantly different WTP distributions (p-value=0.307). However, they reject the null
hypothesis of similar WTP distributions of the OE and the CPC formats (p-value=0.006): the
OE leads to lower values than the CPC. The hypothesis of similar WTP with the PC and the
CPC formats is also rejected (PC leading to lower values than CPC, p-value=0.042).
(PLEASE INSERT FIGURE 3 ABOUT HERE]
15
Overall, the unconditional analyses provide evidence that the CPC format yields higher mean
and median WTP and is less prone to extreme values or anchoring on starting values. However,
because these differences could be due to heterogeneity in respondents’ characteristics across
elicitation formats, in particular regarding income and place of residence (Rural and
Disadvantaged Governorate), we now use interval regression models to control for observed
heterogeneity.
4.3. Comparing formats: Conditional analyses
4.3.1. Separate estimations by elicitation format
For each elicitation format, we look for the determinants of WTP with parsimonious models
obtained by removing variables step by step, starting from the full models and decreasing p-
values (see models 1-3 in Table 3)
[PLEASE INSERT TABLE 3 ABOUT HERE]
In each of the formats, we find a very significant and positive effect of income on WTP - which
argues for the validity of the stated preference survey (see Bishop and Woodward 1995) - and
evidence of an interviewer effect (with a negative effect for interviewers 2 and 4).8 Being
employed or self-employed (Work) has a significant and positive effect on WTP for OE and
CPC.
8 As already mentioned, the interviewer effect also stands for a joint spatial effect. Models with the three spatially
related variables instead of the interviewer dummies show no improvement, although the variables Disadvantaged
Governorate and Public Squares are generally positive and significant.
16
Some variables are significant only for one elicitation format, generally with the expected sign
on WTP: positive for Age, RiskAverse and Married and negative for Elementary, NonDeclared
and Smoker.
4.3.2. Joint estimation over the three elicitation formats
We study the impact of the elicitation format on WTP over the whole sample by introducing
dummy variables for the OE and the PC formats (the reference being the CPC format). Model
4 in Table 3 contains results indicating a significant and negative effect of the OE and the PC
formats, thus confirming the results of the conditional analyses.9
Regarding the determinants of WTP, the joint estimation confirms previous results: an
interviewer effect (with a negative effect for interviewers 2 and 4), a significant and positive
effect of income, Age, Work and RiskAverse on WTP, and a significant and negative effect of
Elementary and NonDeclared.
We finally estimate this model independently on each of the three elicitation formats and test
for joint equality of estimates of each variable over the different formats (see last column in
Table 3). The overall equality test leads to rejection (p-value=0.0385), because of significant
differences between PC and CPC estimates (p-value=0.0089) but not between OE and PC (p-
value=0.4128) nor OE and CPC (p-value=0.3175).
5. DISCUSSION
In assessing the efficacy of a new type of payment card, the CPC, this paper highlights the
impact of the elicitation format on stated WTP in CV surveys. Conditional and unconditional
9 Various models with different sets of control variables (survey-specific, socio-demographic, specific to the health
respondent, specific to the health of the members of the respondent’s family and others, like respondent’s risk
aversion and reasons for not yet having a health insurance scheme) were also estimated as a robustness check and
confirm the results (details upon request).
17
analyses show that the OE and standard PC formats lead to significantly lower values than the
CPC. This provides evidence that respondents may rely on different heuristic decisions when
giving WTP in the OE and in the two PC formats (Hanemann 1996; Welsh and Poe 1998; Frör
2008). This may stem from the fact that, faced with an OE question that is not typical of
purchasing decisions (i.e. setting the price), respondents need to reflect deeply before giving an
amount. The CPC is shown to offer certain advantages: it has better incentive compatibility
properties and it seems less prone to extreme values or anchoring on starting values than the
PC and OE. Moreover, it is not found more difficult by respondents and does not require more
completion time than OE and PC. Consequently, the CPC format may ease WTP elicitation,
especially for socially disadvantaged and poorly educated populations. Should our results be
confirmed by future studies, the CPC may be a useful alternative to already existing formats
for eliciting not only monetary values but any continuous outcome, like discount rate, quality
of life or time preference (van de Pol and Cairns, 2008).
Our analysis has several limitations. First, we do not study the criterion validity of the CPC -
which is rarely done in the literature, and was impossible here, as VHIS for the uninsured does
not exist yet - but only its convergent validity, a less definitive test. The next step should be to
determine whether WTPs elicited with the CPC for traded goods or services come closer to
actual payments than other formats. Second, we focus on differences across elicitation formats
and purposely do not consider protest answers nor true zero WTP, because respondents
answered the first willingness-to-join question before the elicitation format was used. However,
taking self-selection and true zero WTP issues into account should be part of any WTP
modeling aimed at predicting WTP in the Tunisian population. This will be addressed by future
research.
18
Acknowledgments
This work was completed thanks to the support of the A*MIDEX project (no. ANR-11-IDEX-
0001-02) funded by the “Investissements d’Avenir” French Government program, managed by
the French National Research Agency (ANR). The authors are solely responsible for any
remaining errors.
References
Abu-Zaineh, M., Romdhane, H. B., Ventelou, B., Moatti, J.-P., & Chokri, A. (2013).
Appraising financial protection in health: the case of Tunisia. International Journal of
Health Care Finance and Economics, 13(1), 73–93.
Anderson, L. R., & Mellor, J. M. (2009). Are risk preferences stable? Comparing an
experimental measure with a validated survey-based measure. Journal of Risk and
Uncertainty, 39, 137–160.
Arrow, K., Solow, R., Portney, P. R., Leamer, E. E., Radner, R., & Schuman, H. (1993). Report
of the NOAA panel on contingent valuation. Federal Register, 58(10), 4601-4614.
Atkinson, A. B., Rainwater, L., Smeeding, T.M. (1995). Income distribution in OECD
countries, evidence from the Luxembourg income study. Volume 18 of OECD Social Policy
Studies. Paris: Organization for Economic Co-operation and Development.
Barsky, R. B., Juster, F. T., Kimball, M. S., & Shapiro, M. D. (1997). Preference parameters
and behavioral heterogeneity: an experimental approach in the health and retirement study.
The Quarterly Journal of Economics, 112(2), 537-579.
Bateman, I. J., Day, B. H., Dupont, D., & Georgiou, S. (2009). Procedural invariance testing of
the one-and-one-half-bound dichotomous choice elicitation method. The Review of
Economics and Statistics, 91(4), 806-820.
Bateman, I. J., Langford, I. H., Jones, A. P., & Kerr, G.N. (2001). Bound and path effects in
19
double and triple bounded dichotomous choice contingent valuation. Resource and Energy
Economics, 23, 191–214.
Bishop, R., & Woodward, R. (1995). Valuation of environmental amenities under certainty. In
D. Bromley (Ed.), Handbook of Environmental Economics (pp. 543–667). Oxford:
Blackwell Publishers.
Boyle, K., Johnson, F.R., & McCollum, D. (1997). Anchor and adjustment in single-bounded
dichotomous-choice questions. American Journal of Agricultural Economics, 79(5), 1495-
1500.
Cameron, T. A., & Huppert, D. D. (1989). OLS versus ML estimation of non-market resource
values with payment card interval data. Journal of Environmental Economics and
Management, 17, 230-246
Cameron, T. A., Poe, G. L., Ethier, R. G., & Schulze, W. D. (2002). Alternative non-market
value-elicitation methods: are the underlying preferences the same? Journal of
Environmental Economics and Management, 44, 391–425.
Carson, R. T., & Groves, T. (2007). Incentive and information properties of preference
questions. Environmental and Resource Economics, 37, 181‐ 210.
Carson, R. T., & Groves, T. (2011). Incentive and information properties of preference
questions commentary and extensions. In J. Bennett (Ed.), International Handbook on Non‐
market Valuation. Northampton: Edward Elgar.
Carthy, T., Chilton, S., Covey, J., Hopkins, L., Jones-Lee, M., Loomes, G., et al. (1999). On
the contingent valuation of safety and the safety of contingent valuation: Part 2 - the CV/SG
"chained" approach. Journal of Risk and Uncertainty, 17, 187-213.
Chakravarti, I. M., Laha, R. G., & Roy, J. (Eds.) (1967). Handbook of Methods of Applied
Statistics. Vol. 1: Techniques of computation, descriptive methods and statistical inference.
20
London: John Wiley and Sons.
Champ, P. A., & Bishop, R. C. (2006). Is willingness to pay for a public good sensitive to the
elicitation format? Land Economics, 82(2), 162-173.
Cook, J., Jeuland, M., Maskery, B., & Whittington, D. (2012). Giving stated preference
respondents “time to think”: results from four countries. Environmental and Resource
Economics, 51(4), 473-496.
Cooper, J. C., Hanemann, W. M., & Signorello, G. (2002). One-and-one-half-bound
dichotomous choice contingent valuation. Review of Economics and Statistics, 84, 742–750.
Covey, J., Loomes, G., & Bateman, I. (2007). Valuing risk reductions: testing for range biases
in payment card and random card sorting methods. Journal of Environmental Planning and
Management, 50(4), 467-482.
Donaldson, C., Thomas, R., & Torgerson, D. J. (1997). Validity of open-ended and payment
scale approaches to eliciting willingness to pay. Applied Economics, 29(1), 79-84.
Dubourg, W. B., Jones‐ Lee, M., & Loomes, G. (1994). Imprecise preferences and the WTP-
WTA disparity. Journal of Risk and Uncertainty 9: 115-133.
Dubourg, W. B., Jones‐ Lee, M., & Loomes, G. (1997). Imprecise preferences and survey
design in contingent valuation. Economica, 64, 681–702.
Frör, O. (2008). Bounded rationality in contingent valuation: empirical evidence using
cognitive psychology. Ecological Economics, 68, 570-581.
Gatti, R., Angel-Urdinola, D., Silva, J., & Bodor, A. (2011). Striving for better jobs: the
challenge of informality in the Middle East and North Africa region: Overview. Washington,
DC: World Bank.
21
Gibbard, A. (1973). Manipulation of voting schemes: a general result. Econometrica, 41, 587–
601.
Green, D., Jacowitz, K., Kahneman, D., & McFadden, D. (1998). Referendum contingent
valuation, anchoring, and willingness to pay for public goods. Resource and Energy
Economics, 20, 85-116.
Gyrd-Hansen, D., Jensen, M. L., & Kjaer, T. (2014). Framing the willingness-to-pay question:
impact on response patterns and mean willingness to pay. Health Economics, 23, 550–563.
Hammerschmidt, T., Zeitler, H.-P., & Leidl, R. (2003). Unexpected yes- and no-answering
behaviour in the discrete choice approach to elicit willingness to pay: A methodological
comparison with payment cards. International Journal of Health Care Finance and
Economics, 3, 147–166.
Hanemann, W. M. (1996). Theory versus data in the contingent valuation debate. In. J.
Bjornstad, J. R. Kahn J. R. (Eds.), The contingent valuation of environmental resources:
methodological issues and research needs. Brookfield: Elgar.
Herriges, J., & Shogren, J. (1996). Starting point bias in dichotomous choice valuation with
follow-up questioning. Journal of Environmental Economics and Management, 30, 112-131.
Hoehn, J., & Randall, A. (1987). Satisfactory benefit cost indicator from contingent valuation.
Journal of Environmental Economics and Management, 14, 226-247.
Kanninen, B. (1995). Bias in discrete response contingent valuation. Journal of Environmental
Economics and Management, 28, 114-125.
Loayza, N., & Wada, T. (2009). Informal labor in the Middle East and North Africa: basic
Measures and determinants. Washington, D.C.: World Bank.
Lopez-Martin, B., Montes, C., & Benayas, J. (2008). Economic valuation of biodiversity
22
conservation: the meaning of numbers. Conservation Biology, 22, 624–635.
Lunander, A. (1998). Inducing incentives to understate and to overstate willingness to pay
within the open-ended and the dichotomous-choice elicitation formats: an experimental
study. Journal of Environmental Economics and Management, 35, 88-102.
Mahieu, P.-A., Wolff, F.-C., & Shogren, J. (2014). Interval bidding in a distribution elicitation
format. FAERE Working Paper, 2014.16.
Makhloufi, K., Ventelou, B., & Abu-Zaineh, M. (2015). Have health insurance reforms in
Tunisia attained their intended objectives? International Journal of Health Economics and
Management, 15(1), 29-51.
Mitchell, R.-C., & Carson, R. T. (1981). An experiment in determining willingness to pay for
national water quality improvements. US Environmental Protection Agency report (grant
#R 806906010). Washington, D.C.: Resources for the Future.
Mitchell, R. C., & Carson, R. T. (1989). Using surveys to value public goods: the contingent
valuation method. Baltimore: Johns Hopkins University Press.
NIS, National Institute of Statistics Tunisia. (2011). First Results of Employment - National
Survey second quarter, NIS: Tunis.
O’Connell, T., Rasanathan, K., & Chopra, M. (2014). What does universal health coverage
mean? The Lancet, 383(9913), 277–279.
Onwujekwe, O. (2004). Criterion and content validity of a novel structured haggling contingent
valuation question format versus the bidding game and binary with follow-up format. Social
Science & Medicine, 58, 525–537.
Riera, P. (2003). Incentive compatibility in environmental valuation: some positive results,
mimeo presented at EAERE, Bilbao.
23
Ryan, M., Scott, D. A., & Donaldson, C. (2004). Valuing health care using willingness to pay:
a comparison of the payment card and dichotomous choice methods. Journal of Health
Economics, 23(2), 237-258.
Satterthwaite, M. (1975). Strategy-proofness and arrow conditions: existence and
correspondence theorems for voting procedures and welfare functions. Journal of Economic
Theory, 10, 187–217.
Schläpfer, F., & Bräuer, I. (2007). Theoretical incentive properties of contingent valuation
questions: do they matter in the field? Ecological Economics, 62, 451-460.
Shono, A., Kondo, M., Ohmae, H., & Okubo, I. (2014). Willingness to pay for public health
services in rural Central Java, Indonesia: Methodological considerations when using the
contingent valuation method. Social Science & Medicine, 110, 31-40.
Smith R. D. (2006). It's not just what you do, it's the way that you do it: the effect of different
payment card formats and survey administration on willingness to pay for health gain,
Health Economics, 15(3), 281-293.
Uehleke, R. (2014). Willingness to pay for national climate change mitigation policies:
elicitation format and scope effects, mimeo presented at 5th WCERE, Istanbul.
van der Pol, M., & Cairns, J. (2008). Comparison of two methods of eliciting time preference
for future health states. Social Science & Medicine, 67, 883–889
Vega, J. (2013). Universal health coverage: the post-2015 development agenda. Lancet, 381,
179-80.
Veisten, K., & Navrud, S. (2006). Contingent valuation and actual payment for voluntarily
provided passive-use values: assessing the effect of an induced truth-telling mechanism and
elicitation formats. Applied Economics, 38, 735-756.
24
Wang, H., & Whittington, D. (2005). Measuring individuals’ valuation distributions using a
stochastic payment card approach. Ecological Economics, 55, 143-154.
Welsh, M. P., & Poe, G. L. (1998). Elicitation effects in contingent valuation: comparisons to
a multiple bounded discrete choice approach. Journal of Environmental Economics and
Management, 36, 170-185.
World Health Organization (WHO). (2013). The World Health Report 2013: Research for
Universal Health Coverage. Geneva: Switzerland, World Health Organization.
Whynes, D. K., Frew, E., & Wolstenholme, J. L. (2003). A comparison of two methods for
eliciting contingent valuations of colorectal cancer screening. Journal of Health Economics,
22, 555–574.
Whynes, D. K., Frew, E., & Wolstenholme, J. L. (2005). Willingness-to-pay and demand
curves: a comparison of results obtained using different elicitation formats. International
Journal of Health Care Finance and Economics, 5, 369–386.
Whynes, D. .K, Wolstenholme, J. L., & Frew, E. (2004). Evidence of range bias in contingent
valuation payment scales. Health Economics, 13, 183-190.
Wilson, T. D., Houston, C. E., Etling, K. M., & Brekke, N. (1996). A new look at anchoring
effects: basic anchoring and its antecedents. Journal of Experimental Psychology: General,
125(4), 387-402.
Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data (2nd ed.).
Cambridge: MIT Press.
Yang, S.-H., Hu, W., Mupandawana M., & Liu, Y. (2012). Consumer willingness to pay for
fair trade coffee: a Chinese case study. Journal of Agricultural and Applied Economics,
44(1), 21–34.
25
Table 1: Descriptive statistics (n= 426)
Variable definition Mean (Std. Dev.)
Dependent variables
WTP = WTP for VHIS (quarterly, in TND) 42.583 (25.338)a
Respondent characteristics
Male = 1 if male, 0 if female 0.669 (0 .471)
Age = individual’s age (in years) 35.385 (10.395)
Household size = number of household members 2.599 (2.012)
Child = 1 if at least one child under 5 years old in the household, 0 otherwise 0.134 (0.341)
Elderly = 1 if one person over 65 years old in the household, 0 otherwise 0.052 (0.222)
Married = 1 if Married, 0 otherwise 0.417 (0.493)
Illiterate = 1 No schooling, 0 otherwise 0.023 (0.151)
Elementary = 1 primary school, 0 otherwise 0.213 (0.410)
Secondary = 1 secondary education, 0 otherwise 0.516 (0.500)
High School = 1 higher education, 0 otherwise 0.246 (0.431)
Income = Monthly household income (in TND) 558.11 (464.15)
Equivalised Incomeb (Household size)0.5 (in TND) 425.80 (425.72)
Work = 1 if employed /self-employed, 0 otherwise 0.789 (0.409)
Rural = 1 living in rural area, 0 otherwise 0.197 (0.398)
Disadvantaged_gov.c = 1 living in disadvantaged governorate, 0 otherwise 0.443 (0.497)
Other variables
NonDeclared = 1 uninsured due to no declared work, 0 otherwise 0.490 (0.500)
Administration = 1 uninsured due to administrative procedures, 0 otherwise 0.340 (0.474)
NoNeed = 1 uninsured due to no need, 0 otherwise 0.663 (0.198)
RiskAverse = 1 if risk-averse, 0 otherwise d 0.885 (0.319)
Respondent-specific health variables
Self-reported health status = 1 if self-reported health status is good, 0
otherwise
0.835 (0.371)
Outpatient respondent = 1 if at least one outpatient care during the last 3
months, 0 otherwise
0.380 (0.486)
Inpatient respondent = 1 if at least one hospitalization during the last 8
months, 0 otherwise
0.093 (0.292)
Chronic condition = 1 if respondent reports a chronic condition,
0 otherwise
0.124 (0.330)
FinancialHealth = 1 if can afford health services, 0 otherwise 0.370 (0.483)
Smoker = 1 if consuming tobacco products, 0 otherwise 0.460 (0.498)
Health variables specific to the family members of the resp.
Outpatient member = 1 if at least one outpatient care in household during the
last 3 months, 0 otherwise
0.5 (0.500)
Inpatient member = 1 if at least one hospitalization in household during the
last 8 months, 0 otherwise
0.140 (0.348)
Chronic condition = 1 if one household member reports a chronic condition, 0
otherwise
0.185 (0.389)
Survey specific variables
PublicSquare = 1 if sample point is a public square, 0 if informal market 0.420 (0.494)
Interviewer#1-5 = Dummy variables for each of the 5 interviewers -
Time taken to answer the survey (in minutes) 20.427 (3.000)
Proportion that declares difficulties in answering WTP (in %) 0.4137 (.4932) a Ninety respondents refuse to give a WTP. b Equivalised income is computed based on the OECD equivalence scale, by dividing household
income by the square root of household size (see Atkinson et al. 1995).
26
c According to decree n° 2008-387 of February 11, 2008. d Based on six modalities generated according to the method of Barsky et al. (1997).
27
Table 2 Summary statistics by elicitation format and equality tests between elicitation formats
Elicitation formats OE (n=107) PC (n=104) CPC (n=125) All (n=336)
Mean survey completion time [SD]
P-values for equality tests
20.393 [2.787] 20.5 [3.131] 20.395 [3.086]
OE not different from PC (0.7924), PC not different from
CPC (0.7999) and OE not different from CPC (0.9946)
20.427 [3.000]
(0.9564)
Declare difficulty with WTP questions [SD]
P-values for equality tests
0.4112 [0.4944] 0.4519 [0.5001] 0.384 [0.4883]
OE not different from PC (0.5506), PC not different from
CPC (0.2989) and OE not different from CPC (0.6727)
0.4137 [.4932]
(0.584)
Mean WTP [SD] in TND
Median WTP in TND
41.15 [26.64]
40
38.37 [22.10]
35
47.32 [26.12]
45
42.58 [25.34]
40
Equality tests on mean WTP OE not different from PC (0.4092), CPC>PC (0.0054***)
and CPC>OE 0(.0774)*
(0.0220**)
Equality tests on median WTP OE not different from PC (0.493), CPC>PC (0.009***)
and CPC>OE (0.035*)
(0.014**)
Proportion equality test of bid in TND 0-10
range
OE not different from PC (0.2287), PC not different from
CPC (0.1912) and OE>CPC (0.0120**)
(0.0406**)
Proportion equality test of bid over TND 130
range
OE not different from PC (0.1632), PC not different from
CPC (0.3630) and OE not different from CPC (0.4721)
(0.3496)
NB: Outcome of equality tests with corresponding p-values in parentheses. ** if p<0.05, *** if p<0.01. OE=Open-ended, PC=Payment card, CPC=Circular payment card.
28
Table 3 Parsimonious models that explain WTP, by elicitation format
Model 1 (OE) Model 2 (PC) Model 3 (CPC) Model 4 (All) Joint equality test
based on model 4
when estimated by
elicitation format
Variables
OE format (=1) - - - -6.140** (.023)
PC format (=1) - - - -6.821*** (.006)
Survey
Interviewer #2 (=1) -22.243*** (<.0001) -29.221***(<.0001) -10.008*** (.005) -20.587*** (<.0001) (.1866)
Interviewer #4 (=1) -12.941*** (.010) -11.797** (.020) - -6. 208** (.014) (.7097)
Interviewer #3 (=1) - -14.385*** (.001) 9.919** (.045) - -
Socio demo
Equivalised Income .0228*** (<.0001) .0201**(.025) .0255*** (<.0001) .0243*** (<.0001) (.8663)
Work (=1) 10.966** (.014) - 11.463*** (.001) 6.896*** (.016) (.2663)
Age - - .479*** (.001) .2758** (.046) (.0445)
Elementary (=1) - -9.677** (.018) - -5.720* (.058) (.5330)
Other
RiskAverse (=1) 13.773*** (.006) - - 9.398*** (.008) (.3916)
NonDeclared (=1) - - -11.971*** (<.0001) -5.626** (.014) (.0149)
Married (=1) 9.050* (.081) - - - -
Health respondent
Smoking (=1) - -6.513* (.059) - -
Constant 12.471** (.041) 46.957*** (<.0001) 16.616*** (.007) 22.289*** (<.0001) -
Sigma 22.409*** (<.0001) 17.213*** (<.0001) 17.404*** (<.0001) 20.222*** (<.0001) -
Observations 107 104 125 336 107+104+125
LR test of joint nullity 36.947*** (<.0001) 48.701*** (<.0001) 92.524*** (<.0001) 144.814*** (<.0001) 27.27** (.0385)
P-values in parentheses * if p<0.10, ** if p<0.05, *** if p<0.01, OE=Open-ended, PC=Payment card, CPC=Circular payment card.
29
Figure 2 WTP distribution by elicitation format
Figure 3 Proportion of accepted bids by elicitation format
30
Appendix 1: Hypothetical scenario
Introduction by Interviewer:
No one is safe from injury or illness. It is very important to evaluate WTP, because this will
allow a new Voluntary Health Insurance Scheme (VHIS) to be implemented. The VHIS covers
only healthcare benefits and is not conditional on exercising a professional activity (employed
or self-employed). It covers healthcare expenditures of the party insured and his household
members, only with public sector providers. It offers a package of healthcare services exactly
the same as those offered by the public scheme currently run by ‘CNAM’.
The voluntary public scheme will cover beneficiaries for ambulatory and hospitalization health
care services that are provided by the public health sector facilities, with co-payment of 10%
of the total healthcare cost.
We will now ask you questions on the amount that you are willing-to-pay to join this new
voluntary scheme. The value that you are willing-to-pay represents the importance that you
attach to the health insurance scheme and to healthcare services in general. Please note that
this amount will reduce your expenditure on other items.
Note to the interviewer: (Please give the interviewee the blue list that describes the scheme
under consideration and ask her/him to take time to reply to all the questions]
Similar to the blue color of the public scheme currently run by the ‘CNAM’,
these healthcare services are described in this blue list: Exoneration from out-
of-pocket payment in the event of chronic diseases, births, surgical operations,
scanner, medical imaging, hemodialysis, lithotripsy, radiological tests,
physiotherapy, orthopedic and thermal healthcare services.
31
Appendix 2: WTP elicitation questions
Open-Ended elicitation format
Please answer the following question after carefully reading the scheme description.
The insured will pay their contributions quarterly. Considering your income, what is the
maximum amount (in TND) that you are willing-to-pay quarterly to join the scheme under
consideration?
Note to the interviewer: (Please fill in the maximum willing-to-pay amount]
/______________/ TND
Payment card elicitation format
Please answer the following question after carefully reading the scheme description.
The insured will pay their contributions quarterly. Of the amounts illustrated on the payment
card and considering your income, what is the maximum amount (in TND) that you are
willing-to-pay quarterly to join the scheme under consideration?
Note to the interviewer: (Present the payment card] (see below)
0 to 10 TND
11 to 20 TND
21 to 30 TND
31 to 40 TND
41 to 50 TND
51 to 60 TND
61 to 70 TND
71 to 80 TND
81 to 90 TND
91 to 100 TND
101 to 110 TND
111 to 120 TND
121 to 130 TND
> 130 TND
If more than TND 130, please fill in
the exact amount
/______________/ TND
Please tick (√) the interval you are
sure you are willing to pay. Please put a cross (X) against the
interval you are sure you are not
willing to pay.
Please circle (O) the interval of the
maximum amount you are willing
to pay.
32
Circular payment card elicitation format
Please answer the following question after carefully reading the scheme description.
The insured will pay their contributions quarterly. Of the amounts illustrated on the circular
payment card and considering your income, what is the maximum amount (in TND) that you
are willing-to-pay quarterly to join the scheme under consideration?
Note to the interviewer: (Present the circular payment card] (see Figure 1 below)
Please circle (O) the section with the maximum amount you are willing to pay.
Figure 1 Circular payment card (translated into English)